CN113034602A - Orientation angle analysis method and device, electronic equipment and storage medium - Google Patents
Orientation angle analysis method and device, electronic equipment and storage medium Download PDFInfo
- Publication number
- CN113034602A CN113034602A CN202110415929.0A CN202110415929A CN113034602A CN 113034602 A CN113034602 A CN 113034602A CN 202110415929 A CN202110415929 A CN 202110415929A CN 113034602 A CN113034602 A CN 113034602A
- Authority
- CN
- China
- Prior art keywords
- interval
- image
- head
- analyzed
- angle
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/70—Determining position or orientation of objects or cameras
- G06T7/77—Determining position or orientation of objects or cameras using statistical methods
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/161—Detection; Localisation; Normalisation
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20076—Probabilistic image processing
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30196—Human being; Person
- G06T2207/30201—Face
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Oral & Maxillofacial Surgery (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Probability & Statistics with Applications (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
The application provides an orientation angle analysis method and device, electronic equipment and a storage medium, which are used for solving the problem that the obtained orientation angle of a head is inaccurate due to the fact that head posture information for performing head posture estimation on an image is not stable enough. The method comprises the following steps: obtaining an image to be analyzed, and cutting out a head area image in the image to be analyzed; extracting head posture information in the head region image, wherein the head posture information is probability distribution of three components of Euler angles of the face orientation in a preset angle interval; determining the interval screened out at two sides of the maximum probability value in a preset angle interval as a neighborhood interval aiming at each component in the three components of the Euler angle, and calculating the sum of probability distribution in the neighborhood interval to obtain the expected value of each component; and determining the head orientation angle in the image to be analyzed according to the expected value of each component in the three Euler angles.
Description
Technical Field
The present disclosure relates to the field of machine learning, computer vision, and face recognition technologies, and in particular, to a method and an apparatus for analyzing an orientation angle, an electronic device, and a storage medium.
Background
Head pose estimation is one of the popular research directions in the field of computer vision, and refers to predicting pose information of a human head in space through an algorithm, and the head pose estimation is usually a regression problem with a yaw angle (yaw), a pitch angle (pitch), and a roll angle (roll) describing a head pose as a regression target.
In the current image recognition field, head pose estimation is usually required, that is, head pose information of a front direction faced by a human face in a three-dimensional space coordinate is obtained, and an euler angle or a quaternion is usually used; among them, quaternions are often difficult to interpret and unintuitive because they are expressed in a nonlinear relationship with visual perception, and therefore, when intuitive and highly interpretable properties are required, euler's angle is often selected to represent head posture information. However, it has been found during the specific use of euler angles that the head pose information for head pose estimation of an image is not stable enough to result in an inaccurate head orientation angle being obtained.
Disclosure of Invention
An object of the embodiments of the present application is to provide an orientation angle analysis method, an orientation angle analysis device, an electronic device, and a storage medium, which are used to solve the problem that an obtained head orientation angle is inaccurate due to unstable head pose information for performing head pose estimation on an image.
The embodiment of the application provides a method for analyzing an orientation angle, which comprises the following steps: obtaining an image to be analyzed, and cutting out a head area image in the image to be analyzed; extracting head posture information in the head region image, wherein the head posture information is probability distribution of three components of Euler angles of the face orientation in a preset angle interval; determining the interval screened out at two sides of the maximum probability value in a preset angle interval as a neighborhood interval aiming at each component in the three components of the Euler angle, and calculating the sum of probability distribution in the neighborhood interval to obtain the expected value of each component; and determining the head orientation angle in the image to be analyzed according to the expected value of each component in the three Euler angles. In the implementation process, the expected value of each component is determined according to the sum of probability distributions in the interval with the maximum probability value in the preset angle interval extending to the two sides, the angle fitting problem is effectively converted into the interval probability distribution problem, the stability of the head posture information is prevented from being influenced by the characteristics of uneven distribution, obvious truncation and the like of the probability distribution, and therefore the accuracy of obtaining the head orientation angle by the three components of the Euler angle is improved.
Optionally, in this embodiment of the application, determining an interval screened from two sides of the maximum probability value in the preset angle interval as a neighborhood interval includes: dividing a preset angle interval into a plurality of segmentation intervals, and screening a first segmentation interval with the maximum probability value from the plurality of segmentation intervals; and screening a preset number of second sectional intervals from the plurality of sectional intervals according to the directions of two sides of the first sectional interval, and determining the first sectional interval and the preset number of second sectional intervals as neighborhood intervals. In the implementation process, the first subsection interval with the maximum probability and the second subsection intervals with the preset number are determined to be the neighborhood intervals according to the jungle competition rule, so that the influence of the characteristics of uneven distribution, obvious truncation and the like of probability distribution on the stability of the head posture information is avoided, and the stability and the accuracy of obtaining the head orientation angle by the three components of the Euler angle are effectively improved.
Optionally, in this embodiment of the present application, calculating a sum of probability distributions in a neighborhood region to obtain an expected value of each component, includes: normalizing all the segmented intervals in the neighborhood interval, and calculating the median of each segmented interval according to the maximum value and the minimum value of each segmented interval in the neighborhood interval; and determining the sum of the products of the median of each subsection interval in the neighborhood interval and each normalized subsection interval as the expected value of each component.
Optionally, in this embodiment of the present application, cropping out a head region image in an image to be analyzed includes: judging whether a head region in an image to be analyzed is detected; if yes, cutting out a head area image from the image to be analyzed. In the implementation process, the head region cut out in the image to be analyzed is detected; therefore, the quality of the input image is effectively improved, other background in the image is prevented from interfering the recognition result, and the accuracy of obtaining the head orientation angle by the three components of the Euler angle is effectively improved.
Optionally, in this embodiment of the present application, extracting head pose information in the head region image includes: and extracting head posture information in the head region image by using a pre-trained convolutional neural network model.
Optionally, in this embodiment of the present application, before extracting head pose information in the head region image using a pre-trained convolutional neural network model, the method further includes: acquiring a plurality of sample images and a plurality of posture information, wherein the posture information is head posture information of a head area image in the sample images; and training the convolutional neural network by taking the plurality of sample images as training data and the plurality of posture information as training labels to obtain a convolutional neural network model.
Optionally, in an embodiment of the present application, obtaining an image to be analyzed includes: receiving an image to be analyzed, which is acquired by a camera of a truck cockpit; after determining the head orientation angle in the image to be analyzed according to the expected value of each of the three euler angles, the method further comprises the following steps: and if the head orientation angle deviates from the preset range and lasts for a preset duration, generating and outputting early warning information, wherein the early warning information is used for reminding a driver in the truck cab of fatigue driving. In the implementation process, the images to be analyzed collected by the camera of the truck cockpit are received; after the head orientation angle in the image to be analyzed is determined according to the expected value of each component of the three euler angles, the driver in the truck cab is reminded of driving fatigue under the condition that the head orientation angle deviates from the preset range and lasts for the preset duration, so that the probability of safety accidents is reduced, and the range of the application scene of the orientation angle analysis is effectively improved.
The embodiment of the present application further provides an orientation angle analysis device, including: the analysis image obtaining module is used for obtaining an image to be analyzed and cutting out a head area image in the image to be analyzed; the head posture information extraction module is used for extracting head posture information in the head region image, and the head posture information is probability distribution of three components of Euler angles of the face orientation in a preset angle interval; the component expectation obtaining module is used for determining the interval screened out at two sides of the maximum probability value in the preset angle interval as a neighborhood interval aiming at each component in the three components of the Euler angle, and calculating the sum of probability distribution in the neighborhood interval to obtain the expectation value of each component; and the orientation angle determining module is used for determining the orientation angle of the head in the image to be analyzed according to the expected value of each component of the three euler angles.
Optionally, in this embodiment of the present application, the component expectation obtaining module includes: the segmentation interval processing module is used for dividing the preset angle interval into a plurality of segmentation intervals and screening out a first segmentation interval with the maximum probability value from the plurality of segmentation intervals; and the neighborhood interval determining module is used for screening a preset number of second segmentation intervals from the plurality of segmentation intervals according to the directions of two sides of the first segmentation interval, and determining the first segmentation intervals and the preset number of second segmentation intervals as neighborhood intervals.
Optionally, in this embodiment of the present application, the component expectation obtaining module further includes: the interval median calculation module is used for normalizing all the subsection intervals in the neighborhood interval and calculating the median of each subsection interval according to the maximum value and the minimum value of each subsection interval in the neighborhood interval; and the component expectation obtaining module is used for determining the sum of the products of the median of each subsection interval in the neighborhood interval and each normalized subsection interval as the expectation value of each component.
Optionally, in an embodiment of the present application, the analysis image obtaining module includes: the head region judging module is used for judging whether a head region in the image to be analyzed is detected; and the head region cutting module is used for cutting out a head region image from the image to be analyzed if the head region in the image to be analyzed is detected.
Optionally, in an embodiment of the present application, the gesture information extraction module includes: and the network model extraction module is used for extracting the head posture information in the head region image by using a pre-trained convolutional neural network model.
Optionally, in this embodiment of the present application, the orientation angle analyzing apparatus further includes: the image posture acquisition module is used for acquiring a plurality of sample images and a plurality of posture information, wherein the posture information is the head posture information of the head area image in the sample images; and the network model training module is used for training the convolutional neural network by taking the plurality of sample images as training data and the plurality of posture information as training labels to obtain the convolutional neural network model.
Optionally, in this embodiment of the present application, the orientation angle analyzing apparatus further includes: the analysis image acquisition module is used for receiving an image to be analyzed, which is acquired by a camera of a truck cockpit; and the early warning information output module is used for generating and outputting early warning information if the head orientation angle deviates from a preset range and lasts for a preset time, and the early warning information is used for reminding a driver in the truck cab of driving fatigue.
An embodiment of the present application further provides an electronic device, including: a processor and a memory, the memory storing processor-executable machine-readable instructions, the machine-readable instructions when executed by the processor performing the method as described above.
Embodiments of the present application also provide a storage medium having a computer program stored thereon, where the computer program is executed by a processor to perform the method as described above.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are required to be used in the embodiments of the present application will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present application and therefore should not be considered as limiting the scope, and that those skilled in the art can also obtain other related drawings based on the drawings without inventive efforts.
FIG. 1 is a schematic diagram illustrating probability distributions corresponding to a pitch angle, a yaw angle and a roll angle provided by an embodiment of the present application;
fig. 2 is a schematic flow chart of an orientation angle analysis method provided in an embodiment of the present application;
FIG. 3 is a flow chart illustrating a method for analyzing an orientation angle in a long-distance freight scenario according to an embodiment of the present application;
fig. 4 is a schematic structural diagram of an orientation angle analysis apparatus provided in an embodiment of the present application;
fig. 5 is a schematic structural diagram of an electronic device provided in an embodiment of the present application.
Detailed Description
Before describing the orientation angle analysis method provided by the embodiments of the present application, some concepts involved in the embodiments of the present application are described below:
deep Learning (Deep Learning), which refers to Learning the intrinsic rules and representation levels of sample data, wherein information obtained in the Learning process is helpful to the interpretation of data such as characters, images and sounds; the final goal of deep learning is to make a machine capable of human-like analytical learning, and recognizing data such as characters, images and sounds, and the deep learning includes, but is not limited to, extracting features of data such as characters, images and sounds by using a deeper neural network model.
A Convolutional Neural Network (CNN), which is an artificial Neural network, in which artificial neurons of the artificial Neural network can respond to surrounding units and can perform large-scale image processing; the convolutional neural network may include convolutional and pooling layers. The convolutional neural network includes a one-dimensional convolutional neural network, a two-dimensional convolutional neural network, and a three-dimensional convolutional neural network.
Face Recognition (Face Recognition), which is a research field of computer technology, belongs to the biological feature Recognition technology and is used for distinguishing organism individuals from biological features of organisms (generally, specially, people); the generalized face recognition actually comprises a series of related technologies for constructing a face recognition system, including face image acquisition, face positioning, face recognition preprocessing, identity confirmation, identity search and the like; the narrow-sense face recognition refers to a technique or system for identity confirmation or identity search through a face.
The jungle competition law refers to a view point on probability distribution theory, namely, the interval with the maximum probability has the reference value and has the leading position, and the farther the interval is from the maximum probability, the smaller the reference value is; the segment interval with the highest probability in the domain interval has the highest reference value, and the farther the segment interval with the highest probability is, the smaller the reference value of the probability value corresponding to the segment interval is.
Before describing the orientation angle analysis method provided by the embodiment of the present application, the following describes problems of a comparison embodiment, in which unstable head pose information obtained by generally directly fitting using euler angles may occur, that is, unstable head pose information for performing head pose estimation on an image includes: the probability distribution of the three components of the Euler angle of the face orientation, such as uneven distribution, abnormal distribution, obvious truncation and the like, of the head posture information; among these, the three components here include: yaw (yaw), pitch (pitch) and roll (roll).
Please refer to fig. 1, which illustrates a schematic diagram of probability distribution corresponding to a pitch angle, a yaw angle, and a roll angle provided in an embodiment of the present application; the vertical coordinates of the three components in the figure represent specific probability values (which can also be understood as frequency coefficients), the corresponding horizontal coordinates are all intervals in which the values are located, namely, angle intervals from negative 99 degrees to positive 99 degrees, and one interval is set every 3 degrees in the figure. When the probability distribution of the three components of the Euler angle is converted into the frequency logarithmic distribution of the three components in an application scene, the frequency logarithmic distribution is unbalanced and has truncation; therefore, if the neural network is trained using these unstable head orientation angles, the stability and accuracy of obtaining the neural network model can have an impact.
It should be noted that the orientation angle analysis method provided in the embodiments of the present application may be executed by an electronic device, where the electronic device refers to a device terminal or a server having a function of executing a computer program, and the device terminal includes, for example: a smart phone, a Personal Computer (PC), a tablet computer, a Personal Digital Assistant (PDA), a Mobile Internet Device (MID), a network switch or a network router, and the like.
Before introducing the orientation angle analysis method provided in the embodiment of the present application, an application scenario applicable to the orientation angle analysis method is introduced, where the application scenario includes, but is not limited to: using the orientation angle analysis method to determine whether a driver in a cockpit of a truck, car, subway train, or train, etc. is inattentive, where inattentive conditions include, but are not limited to: fatigue driving or drunk driving, etc., so as to prompt the driver to concentrate on driving or prompt the driver to start the automatic driving mode in time, so as to replace all or part of driving operation of the driver. Of course, in a specific implementation process, the orientation angle analysis method may also be applied to an artificial intelligence driving system or a vehicle assistant driving system, specifically for example: the function of judging whether the driver focuses attention of an artificial intelligence driving system or a vehicle auxiliary driving system is enhanced by using the orientation angle analysis method.
Please refer to fig. 2, which is a schematic flow chart of an orientation angle analysis method according to an embodiment of the present application; the main idea of the orientation angle analysis method is that the expected value of each component is determined by the sum of probability distributions in an interval which is expanded to two sides according to the maximum probability value in a preset angle interval, the angle fitting problem is effectively converted into the interval probability distribution problem, the influence of characteristics such as uneven distribution, obvious truncation and the like of the probability distribution on the stability of head posture information is avoided, and the accuracy of obtaining the orientation angle of the head by three components of the Euler angle is improved; the orientation angle analysis method may include:
step S110: and obtaining an image to be analyzed, and cutting out a head area image in the image to be analyzed.
The obtaining method of the image to be analyzed in the step S110 includes: in the first acquisition mode, a video camera (for example, a security camera or a home camera) with an RGB channel, a video recorder, a color camera or other terminal equipment is used to shoot a target object to obtain an image to be analyzed; then the terminal device sends an image to be analyzed to the electronic device, then the electronic device receives the image to be analyzed sent by the terminal device, and the electronic device can store the image to be analyzed into a file system, a database or a mobile storage device; the second obtaining method is to obtain a pre-stored image to be analyzed, and specifically includes: acquiring an image to be analyzed from a file system, or acquiring the image to be analyzed from a database, or acquiring the image to be analyzed from a mobile storage device; in the third obtaining mode, software such as a browser is used for obtaining the image to be analyzed on the internet, or other application programs are used for accessing the internet to obtain the image to be analyzed.
The embodiment of cropping out the head region image in the image to be analyzed in step S110 described above is, for example: detecting a head region in an image to be analyzed, and judging whether the head region in the image to be analyzed is detected; if the head area in the image to be analyzed is detected, cutting out a head area image from the image to be analyzed; specifically, a Dual Shot Face Detector (DSFD) or a very small Face Detector (exit Face Detector) and other head detection algorithms can be used to detect the head region in the image to be analyzed; the detection algorithm here may adopt a head detection algorithm, or adopt a face recognition method to locate a face region, and expand the face region to a preset region, so as to obtain the head region, and the adopted face recognition related neural network model includes but is not limited to: multi-task Cascaded Convolutional Neural Networks (MTCNN), Regional Convolutional Neural Networks (RCNN), and so on.
After step S110, step S120 is performed: extracting head posture information in the head region image, wherein the head posture information is probability distribution of three components of Euler angles of the face orientation in a preset angle interval.
There are many embodiments of the above step S120, including but not limited to the following:
in a first implementation manner, a trained deep learning model is directly obtained, and the trained neural network model is used to extract head pose information in a head region image, where the obtaining of the trained deep learning model may include: the first obtaining mode is to receive the deep learning model sent by other terminal equipment and store the deep learning model into a file system, a database or mobile storage equipment; the second obtaining method obtains a pre-stored deep learning model, specifically for example: acquiring a deep learning model from a file system, or acquiring the deep learning model from a database, or acquiring the deep learning model from a mobile storage device; and the third obtaining mode is to obtain the deep learning model on the internet by using software such as a browser, or obtain the deep learning model by accessing the internet by using other application programs.
In a second embodiment, training a neural network model for extracting head pose information from a head region image from the beginning, and extracting head pose information from the head region image using the trained neural network model may include: carrying out image enhancement on training data sets such as LFPW, IBUG, HELEN, AFW and the like commonly used in academia to obtain an image-enhanced training data set, wherein the image enhancement operation which can be adopted comprises the following steps: inversion, rotation, translation, cropping, contrast adjustment, noise addition, image scaling, and the like; and training the deep learning model by using the training data set after image enhancement, and obtaining the deep learning model with stable convergence after full training.
The specific process of model training is as follows: acquiring a plurality of sample images and a plurality of posture information, wherein the posture information is head posture information of a head area image in the sample images; training the convolutional neural network by taking a plurality of sample images as training data and a plurality of posture information as training labels to obtain a convolutional neural network model; the deep learning model herein can adopt convolutional neural network models such as a LeNet network model, an AlexNet network model, a VGG network model, a GoogLeNet network model, a ResNet network model, and the like. After the deep learning model is trained, the head posture information in the head region image is extracted by using the trained deep learning model. In a specific practice process, the training data set may be further divided into a training set and a verification set according to specific situations, for example: the method comprises the steps of enhancing images of a training data set to obtain 68.8 ten thousand samples, selecting 92.5% of the samples from the training data set as the training set to train the deep learning model, using the rest 7.5% of the samples in the training data set as a verification set, and verifying the training effect of the trained deep learning model by using the verification set.
In order to further improve the stability of the result prediction of the neural network model, the convolutional neural network may also be self-supervised trained according to the screened video data, and the embodiment is as follows: before obtaining a plurality of sample images, obtaining a plurality of sample videos, and then screening the plurality of sample videos according to a gradient limiting diagram, wherein the gradient limiting diagram refers to a head posture change degree threshold value in a video, namely screening the sample videos with head posture change gradients smaller than the gradient limiting diagram from the plurality of sample videos, obtaining screened video data, and then extracting the sample images from the screened video data, thereby obtaining the plurality of sample images. The specific process of screening a plurality of sample videos according to the gradient clip map includes: assuming that the interval number of a certain attitude variable is m, k frames of continuous video data can be taken, then an m × k matrix is obtained, 8 pre-defined gradient graphs (namely 8-direction gradient convolution kernels) are used for carrying out gradient calculation on the m × k matrix, then the head attitude change gradient of each frame in the k frames of continuous video data can be obtained, then the head attitude change gradient is compared with a gradient limiting graph, and if the loss value between the head attitude change gradient and the gradient limiting graph is larger than 0, the video data is removed from a plurality of sample videos.
The above-described process of comparing the head pose change gradient with the gradient slice map is, for example: first according to the formula
Calculating a loss value between the head posture change gradient and the gradient amplitude limiting graph; wherein G isi,jIs the jth gradient map of the ith pose,
the gradient map is the ith gradient map in 8 predefined gradient maps; and when the head posture change gradient is larger than the gradient limiting map, the loss value between the head posture change gradient and the gradient limiting map is larger than 0, otherwise, the loss value between the head posture change gradient and the gradient limiting map is set to be 0.
It is understood that the above 8 gradient maps can be specifically defined as: the gradient in the first north direction is shown as
The second gradient map in the northeast direction is
The third orthodontistic gradient map is
The fourth gradient map in the southeast direction is
The fifth positive south gradient map is
The sixth gradient map in the southwest direction is
The seventh gradient map in the Zhengxi direction is
The eighth gradient map in the northwest direction is
In training the neural network model, the problem of head pose estimation (i.e., head orientation angle analysis) can be regarded as a linear regression problem, and a Mean Square Error (MSE) loss function, a cross entropy classification loss function, or an L1 norm loss function (L1 _ loss) is used, and the L1 norm loss function can also be referred to as a Least Absolute Deviation (LAD) loss function. Specifically, a convolutional neural network can be adopted to calculate the picture data in the training data set to obtain predicted head posture information; the distance between the predicted head pose information and the true head pose information (e.g., picture labels in the training dataset) is measured, i.e., "loss is greater for farther distances and less for closer distances".
After step S120, step S130 is performed: and aiming at each component in the three components of the Euler angle, determining the interval screened out at the two sides of the maximum probability value in the preset angle interval as a neighborhood interval, and calculating the sum of probability distribution in the neighborhood interval to obtain the expected value of each component.
The preset angle section refers to a range section of the preset orientation angle of the head or the face, and specifically includes: the head orientation angle of the driver in the truck cabin which needs to be highly concentrated is generally between-99 degrees and +99 degrees, and if the head orientation angle is negative, the head orientation angle indicates leftward deviation, and if the head orientation angle is positive, the head orientation angle indicates leftward deviation, and if the head orientation angle indicates rightward deviation, the head orientation angle is [ -99, +99] here can be understood as the preset angle interval. In a specific implementation process, the total number of intervals may be an odd number, specifically for example: the pitch angle can be divided into 11 intervals, and the interval is 10 deg; dividing the yaw angle into 19 intervals, wherein the interval is 10 deg; the roll angle is divided into 5 intervals, and the interval of the intervals is 5 deg; the range of the pitching attitude interval is-15-95 deg, the range of the yawing attitude interval is-95 deg, and the range of the rolling attitude interval is-22.5 deg.
The above-mentioned embodiment of determining the regions screened from both sides of the maximum probability value in the preset angle region as the neighborhood regions in step S130 may include:
step S131: the preset angle interval is divided into a plurality of segmentation intervals, and a first segmentation interval with the maximum probability value is screened out from the segmentation intervals.
The embodiment of step S131 described above is, for example: the preset angle interval [ -99, +99 [)]The interval is divided into a plurality of segmented intervals according to the mode that each interval is 3 degrees, and the plurality of intervals can be expressed as d by using a formula1,d2,…,dnThen n here is 66, where each interval d isiA probability value p (d) is obtainedi) Of course, the preset angle interval can be set according to the specific situation, for example, to [ -90, +90 [ -90 [ ]]And the like; then, the first segmentation interval with the maximum probability value is screened out from the plurality of segmentation intervals according to the jungle competition rule.
Step S132: and screening a preset number of second sectional intervals from the plurality of sectional intervals according to the directions of two sides of the first sectional interval, and determining the first sectional interval and the preset number of second sectional intervals as neighborhood intervals.
The embodiment of step S132 described above is, for example: the jungle rule considers that the segment interval with the maximum probability has the highest reference value, and the farther the segment interval with the maximum probability is, the smaller the reference value of the probability value corresponding to the segment interval is. Therefore, the neighborhood region with the maximum probability can be adopted for calculation, and the neighborhood region is abandoned at other places. Screening out a preset number of second segmentation intervals from the plurality of segmentation intervals according to the directions of two sides of the first segmentation interval with the maximum probability value, and determining the first segmentation interval with the maximum probability value and the preset number of second segmentation intervals as neighborhood intervals; specific examples thereof include: the span size of the neighborhood interval can be controlled by using a parameter K, and the first segmentation interval with the maximum probability value is used asAnd (K-1)/2 second segmentation intervals are expanded to the left side and the right side of the reference, and then the first segmentation interval with the maximum probability value and the (K-1)/2 second segmentation intervals are determined as the neighborhood intervals. If the neighborhood zone exceeds the zone boundary, the preset angle zone [ -99, +99] is exceeded]The excess portion can be discarded directly. It can be seen that a neighborhood region is composed of a plurality of consecutive segment regions, i.e. the neighborhood region can be expressed as σ ═ d using the formulai,di+1,…,di+n}. The parameter K is an adjustable parameter, and is adjusted according to different application scenarios, for example: the parameter K may be set to an odd number in some scenarios,
the above embodiment of calculating the sum of the probability distributions in the neighborhood zone in step S130 includes:
step S133: all the segmented intervals in the neighborhood interval are normalized, and the median of each segmented interval is calculated according to the maximum value and the minimum value of each segmented interval in the neighborhood interval.
The embodiment of step S133 described above includes, for example: all segment intervals within the neighborhood interval are normalized, i.e. the formula can be expressed as: order to
P (d) ← p (d)/s, wherein d represents one of the segment intervals, σ represents a neighborhood interval, (d ∈ σ) represents a segment interval belonging to the neighborhood interval, p (d) represents a gaussian probability distribution value of the segment interval, and s represents the sum of the gaussian probability distribution values of all the segment intervals in the neighborhood interval; then according to
Calculating the median of each subsection interval; where mid (d) represents the median value of the segmentation interval d, upperbound (d) represents the higher boundary probability value of the left and right boundaries, and lowerbound (d) represents the lower boundary probability value of the left and right boundaries. Of course, in a specific implementation, the above normalization function may also be a softmax function, and therefore, the normalization operation herein is not specifically selected as the normalization functionShould be understood as a limitation of the embodiments of the application.
Step S134: and determining the sum of the products of the median of each subsection interval in the neighborhood interval and each normalized subsection interval as the expected value of each component.
The embodiment of step S134 described above is, for example: according to the formula
To calculate an expected value for each component; where d denotes one of the segment intervals, σ denotes a neighborhood interval, p (d) denotes a probability distribution value of the segment interval, mid (d) denotes a median value of the segment interval d, and Exp denotes an expected value of each component. It is understood that after the steps S131 to S134 are performed for each of the three euler angles, the expected value of each of the three euler angles can be obtained.
After step S130, step S140 is performed: and determining the head orientation angle in the image to be analyzed according to the expected value of each component in the three Euler angles.
The embodiment of step S140 described above is, for example: after obtaining the desired values for each of the three components of euler angle, the desired values for each of the three components of euler angle can be used to determine the head orientation angle in the image to be analyzed.
In a specific implementation, the problem of head pose estimation (i.e., head orientation angle analysis) can be converted into a discrete interval classification problem, and cross entropy or other classification loss functions can be used. In order to reduce the occurrence of the jump phenomenon of the prediction result, that is, in the training process of the convolutional neural network, in order to use the cross entropy classification loss function to embody the characteristic that the loss is larger when the distance is farther away and the loss is smaller when the distance is closer, a gaussian probability branch for performing probability weighting on the corresponding interval of each of the three euler angle components can be added to the output layer of the convolutional neural network, so that the three euler angle components of the output of the convolutional neural network all obey gaussian distribution. The network structure of the gaussian probability branch can adopt a traditional classification branch network structure, namely the output characteristic quantity of the output layer adopts the interval quantity of the head posture information, and the gaussian cross entropy loss can be adopted in the training process of the gaussian probability branch, so that three components of the output euler angle of the convolutional neural network are subjected to gaussian distribution.
In the implementation process, firstly, a head region image in an image to be analyzed is cut out, head posture information including three components of Euler angles in the head region image is extracted, then, an interval with the maximum probability value expanding to two sides in a preset angle interval is determined as a neighborhood interval, the sum of probability distribution in the neighborhood interval is calculated to obtain the expected value of each component, and finally, the head orientation angle in the image to be analyzed is determined according to the expected value of each component in the three components of Euler angles; that is to say, the expected value of each component is determined according to the sum of probability distributions in the interval with the maximum probability value in the preset angle interval expanding to the two sides, the angle fitting problem is effectively converted into the interval probability distribution problem, the stability of the head posture information is prevented from being influenced by the characteristics of uneven distribution, obvious truncation and the like of the probability distribution, and therefore the accuracy of obtaining the head orientation angle by the three components of the euler angle is improved. Please refer to fig. 3, which is a schematic flow chart of an orientation angle analysis method in a long-distance freight scenario according to an embodiment of the present application; optionally, the orientation angle analysis method may also be used in the long-distance freight industry, and of course, may also be applied to other industries, where a long-distance freight scene is taken as an example for description, and specifically may include:
step S210: the electronic equipment receives an image to be analyzed, which is acquired by a camera of a truck cockpit, and cuts out a head area image in the image to be analyzed.
The embodiment of step S210 described above is, for example: the electronic device receives an image to be analyzed collected by a camera of a truck cockpit through a Transmission Control Protocol (TCP) or a User Datagram Protocol (UDP), and cuts out a head area image in the image to be analyzed.
Step S220: the electronic equipment extracts head posture information in the head region image, wherein the head posture information is probability distribution of three components of Euler angles of the face orientation in a preset angle interval.
Step S230: the electronic equipment determines the interval screened out at the two sides of the maximum probability value in the preset angle interval as a neighborhood interval aiming at each component in the three components of the Euler angle, calculates the sum of probability distribution in the neighborhood interval and obtains the expected value of each component.
Step S240: the electronic device determines the head orientation angle in the image to be analyzed from the expected value of each of the three euler angles.
The implementation principle and implementation manner of the above steps S220 to S240 are similar to those of the steps S120 to S140, and therefore, the implementation principle and implementation manner of the steps are not described herein, and if not clear, reference may be made to the description of the steps S120 to S140.
Step S250: and if the head orientation angle deviates from the preset range and lasts for the preset duration, the electronic equipment generates and outputs early warning information, and the early warning information is used for reminding a driver in the truck cab of fatigue driving.
The embodiment of the step S250 is, for example: if the head orientation angle of a driver in the truck cab deviates from a preset range of [ -99, +99], and the head orientation angle lasts for a preset time period of 4 seconds; the preset range and the preset duration may be adjusted according to the actual application scenario, for example, the preset range is set to [ -90, +90], the preset duration is set to 3 seconds or 5 seconds, and so on.
Of course, in a specific implementation process, the orientation angle analysis method may also be applied to an artificial intelligence driving system or a vehicle assistant driving system, specifically for example: the function of judging whether the driver concentrates on attention of an artificial intelligent driving system or a vehicle auxiliary driving system is enhanced by using the orientation angle analysis method; if it is detected that the head orientation angle of the driver deviates from the preset range and the time duration exceeds the preset time, the vehicle auxiliary driving system can be used for assisting the driver in driving (such as forcibly turning a double flashing light, parking beside and the like), or the artificial intelligent driving system can be directly used for taking over the driver to continue driving and the like.
In the implementation process, the images to be analyzed collected by the camera of the truck cockpit are received; after the head orientation angle in the image to be analyzed is determined according to the expected value of each component of the three euler angles, the driver in the truck cab is reminded of driving fatigue under the condition that the head orientation angle deviates from the preset range and lasts for the preset duration, so that the probability of safety accidents is reduced, and the range of the application scene of the orientation angle analysis is effectively improved.
Please refer to fig. 4, which illustrates a schematic structural diagram of an orientation angle analyzing apparatus according to an embodiment of the present application; the embodiment of the present application provides an orientation angle analysis device 300, including:
and an analysis image obtaining module 310, configured to obtain an image to be analyzed, and crop out a head region image in the image to be analyzed.
The pose information extracting module 320 is configured to extract head pose information in the head region image, where the head pose information is probability distribution of three components of euler angles of the face orientation in a preset angle interval.
And a component expectation obtaining module 330, configured to determine, for each component of the three components of the euler angle, an interval screened on both sides of the maximum probability value in the preset angle interval as a neighborhood interval, and calculate a sum of probability distributions in the neighborhood interval to obtain an expected value of each component.
An orientation angle determining module 340, configured to determine a head orientation angle in the image to be analyzed according to the expected value of each of the three euler angles.
Optionally, in this embodiment of the present application, the component expectation obtaining module includes:
and the segmentation interval processing module is used for dividing the preset angle interval into a plurality of segmentation intervals and screening out a first segmentation interval with the maximum probability value from the plurality of segmentation intervals.
And the neighborhood interval determining module is used for screening a preset number of second segmentation intervals from the plurality of segmentation intervals according to the directions of two sides of the first segmentation interval, and determining the first segmentation intervals and the preset number of second segmentation intervals as neighborhood intervals.
Optionally, in this embodiment of the present application, the component expectation obtaining module further includes:
and the interval median calculation module is used for normalizing all the subsection intervals in the neighborhood interval and calculating the median of each subsection interval according to the maximum value and the minimum value of each subsection interval in the neighborhood interval.
And the component expectation obtaining module is used for determining the sum of the products of the median of each subsection interval in the neighborhood interval and each normalized subsection interval as the expectation value of each component.
Optionally, in an embodiment of the present application, the analysis image obtaining module includes:
and the head region judging module is used for judging whether the head region in the image to be analyzed is detected.
And the head region cutting module is used for cutting out a head region image from the image to be analyzed if the head region in the image to be analyzed is detected.
Optionally, in an embodiment of the present application, the gesture information extraction module includes:
and the network model extraction module is used for extracting the head posture information in the head region image by using a pre-trained convolutional neural network model.
Optionally, in this embodiment of the present application, the orientation angle analyzing apparatus further includes:
the image posture acquisition module is used for acquiring a plurality of sample images and a plurality of posture information, wherein the posture information is the head posture information of the head area image in the sample images.
And the network model training module is used for training the convolutional neural network by taking the plurality of sample images as training data and the plurality of posture information as training labels to obtain the convolutional neural network model.
Optionally, in this embodiment of the present application, the orientation angle analyzing apparatus further includes:
and the analysis image acquisition module is used for receiving the image to be analyzed acquired by the camera of the truck cockpit.
And the early warning information output module is used for generating and outputting early warning information if the head orientation angle deviates from a preset range and lasts for a preset time, and the early warning information is used for reminding a driver in the truck cab of driving fatigue.
It should be understood that the apparatus corresponds to the above-mentioned orientation angle analysis method embodiment, and can perform the steps related to the above-mentioned method embodiment, and the specific functions of the apparatus can be referred to the above description, and the detailed description is appropriately omitted here to avoid redundancy. The device includes at least one software function that can be stored in memory in the form of software or firmware (firmware) or solidified in the Operating System (OS) of the device.
Please refer to fig. 5, which illustrates a schematic structural diagram of an electronic device according to an embodiment of the present application. An electronic device 400 provided in an embodiment of the present application includes: a processor 410 and a memory 420, the memory 420 storing machine-readable instructions executable by the processor 410, the machine-readable instructions when executed by the processor 410 performing the method as above.
The embodiment of the present application also provides a storage medium 430, where the storage medium 430 stores a computer program, and the computer program is executed by the processor 410 to perform the method as above.
The storage medium 430 may be implemented by any type of volatile or nonvolatile storage device or combination thereof, such as a Static Random Access Memory (SRAM), an Electrically Erasable Programmable Read-Only Memory (EEPROM), an Erasable Programmable Read-Only Memory (EPROM), a Programmable Read-Only Memory (PROM), a Read-Only Memory (ROM), a magnetic Memory, a flash Memory, a magnetic disk, or an optical disk.
In the embodiments provided in the present application, it should be understood that the disclosed apparatus and method may be implemented in other ways. The apparatus embodiments described above are merely illustrative, and for example, the flowchart and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods and computer program products according to various embodiments of the present application. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.
In addition, functional modules of the embodiments in the present application may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.
In this document, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions.
The above description is only an alternative embodiment of the embodiments of the present application, but the scope of the embodiments of the present application is not limited thereto, and any person skilled in the art can easily conceive of changes or substitutions within the technical scope of the embodiments of the present application, and all the changes or substitutions should be covered by the scope of the embodiments of the present application.
Claims (10)
1. An orientation angle analysis method, comprising:
obtaining an image to be analyzed, and cutting out a head area image in the image to be analyzed;
extracting head posture information in the head region image, wherein the head posture information is probability distribution of three components of Euler angles of the face orientation in a preset angle interval;
determining the interval screened out at two sides of the maximum probability value in the preset angle interval as a neighborhood interval aiming at each component in the three components of the Euler angle, and calculating the sum of probability distribution in the neighborhood interval to obtain the expected value of each component;
and determining the head orientation angle in the image to be analyzed according to the expected value of each component in the three Euler angle components.
2. The method of claim 1, wherein the determining the regions screened from both sides of the maximum probability value in the preset angle region as neighborhood regions comprises:
dividing a preset angle interval into a plurality of segmentation intervals, and screening out a first segmentation interval with the maximum probability value from the plurality of segmentation intervals;
and screening a preset number of second sectional intervals from the plurality of sectional intervals according to the directions of the two sides of the first sectional interval, and determining the first sectional interval and the preset number of second sectional intervals as the neighborhood intervals.
3. The method of claim 1, wherein calculating the sum of the probability distributions in the neighborhood region to obtain the expected value of each component comprises:
normalizing all the segmented intervals in the neighborhood interval, and calculating the median of each segmented interval according to the maximum value and the minimum value of each segmented interval in the neighborhood interval;
and determining the sum of the products of the median of each subsection interval in the neighborhood interval and each normalized subsection interval as the expected value of each component.
4. The method of claim 1, wherein the cropping out the head region image in the image to be analyzed comprises:
judging whether a head region in the image to be analyzed is detected;
and if so, cutting out the head area image from the image to be analyzed.
5. The method according to claim 1, wherein the extracting head pose information in the head region image comprises:
and extracting head posture information in the head region image by using a pre-trained convolutional neural network model.
6. The method of claim 5, further comprising, before the extracting head pose information in the head region image using the pre-trained convolutional neural network model:
acquiring a plurality of sample images and a plurality of posture information, wherein the posture information is head posture information of a head area image in the sample images;
and training a convolutional neural network by taking the sample images as training data and the posture information as training labels to obtain the convolutional neural network model.
7. The method according to any one of claims 1-6, wherein said obtaining an image to be analyzed comprises:
receiving the image to be analyzed acquired by a camera of a truck cockpit;
after determining the head orientation angle in the image to be analyzed according to the expected value of each of the three euler angle components, the method further comprises the following steps:
and if the head orientation angle deviates from a preset range and lasts for a preset duration, generating and outputting early warning information, wherein the early warning information is used for reminding a driver in the truck cab of fatigue driving.
8. An orientation angle analyzing apparatus, comprising:
the analysis image obtaining module is used for obtaining an image to be analyzed and cutting out a head area image in the image to be analyzed;
the head posture information extraction module is used for extracting head posture information in the head region image, and the head posture information is probability distribution of three components of Euler angles of the face orientation in a preset angle interval;
a component expectation obtaining module, configured to determine, for each component of the three components of the euler angle, an interval screened on both sides of the maximum probability value in the preset angle interval as a neighborhood interval, and calculate a sum of probability distributions in the neighborhood interval to obtain an expected value of each component;
and the orientation angle determining module is used for determining the orientation angle of the head in the image to be analyzed according to the expected value of each component in the three euler angle components.
9. An electronic device, comprising: a processor and a memory, the memory storing machine-readable instructions executable by the processor, the machine-readable instructions, when executed by the processor, performing the method of any of claims 1 to 7.
10. A storage medium, having stored thereon a computer program which, when executed by a processor, performs the method of any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110415929.0A CN113034602B (en) | 2021-04-16 | 2021-04-16 | Orientation angle analysis method and device, electronic equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110415929.0A CN113034602B (en) | 2021-04-16 | 2021-04-16 | Orientation angle analysis method and device, electronic equipment and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113034602A true CN113034602A (en) | 2021-06-25 |
| CN113034602B CN113034602B (en) | 2023-04-07 |
Family
ID=76456777
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110415929.0A Active CN113034602B (en) | 2021-04-16 | 2021-04-16 | Orientation angle analysis method and device, electronic equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113034602B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116844210A (en) * | 2023-07-04 | 2023-10-03 | 宁夏大学 | Head posture estimation method and system based on structure isodegeneration self-supervision learning |
| CN117197853A (en) * | 2022-05-31 | 2023-12-08 | 青岛云天励飞科技有限公司 | Face angle prediction method, device, equipment and readable storage medium |
| WO2025180426A1 (en) * | 2024-02-26 | 2025-09-04 | 北京罗克维尔斯科技有限公司 | Face orientation detection method and apparatus, and device, vehicle, medium and program product |
| CN117197853B (en) * | 2022-05-31 | 2025-10-10 | 青岛云天励飞科技有限公司 | Face angle prediction method, device, equipment and readable storage medium |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109255329A (en) * | 2018-09-07 | 2019-01-22 | 百度在线网络技术（北京）有限公司 | Determine method, apparatus, storage medium and the terminal device of head pose |
| WO2019056471A1 (en) * | 2017-09-19 | 2019-03-28 | 平安科技（深圳）有限公司 | Driving model training method, driver recognition method and apparatus, device, and medium |
| WO2019128646A1 (en) * | 2017-12-28 | 2019-07-04 | 深圳励飞科技有限公司 | Face detection method, method and device for training parameters of convolutional neural network, and medium |
| CN110263678A (en) * | 2019-06-03 | 2019-09-20 | 上海眼控科技股份有限公司 | A kind of face direction determination process and system |
| CN112183482A (en) * | 2020-10-29 | 2021-01-05 | 苏州市职业大学 | Dangerous driving behavior identification method, device, system and readable storage medium |
| CN112541436A (en) * | 2020-12-15 | 2021-03-23 | 平安科技（深圳）有限公司 | Concentration degree analysis method and device, electronic equipment and computer storage medium |
-
2021
- 2021-04-16 CN CN202110415929.0A patent/CN113034602B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019056471A1 (en) * | 2017-09-19 | 2019-03-28 | 平安科技（深圳）有限公司 | Driving model training method, driver recognition method and apparatus, device, and medium |
| WO2019128646A1 (en) * | 2017-12-28 | 2019-07-04 | 深圳励飞科技有限公司 | Face detection method, method and device for training parameters of convolutional neural network, and medium |
| CN109255329A (en) * | 2018-09-07 | 2019-01-22 | 百度在线网络技术（北京）有限公司 | Determine method, apparatus, storage medium and the terminal device of head pose |
| CN110263678A (en) * | 2019-06-03 | 2019-09-20 | 上海眼控科技股份有限公司 | A kind of face direction determination process and system |
| CN112183482A (en) * | 2020-10-29 | 2021-01-05 | 苏州市职业大学 | Dangerous driving behavior identification method, device, system and readable storage medium |
| CN112541436A (en) * | 2020-12-15 | 2021-03-23 | 平安科技（深圳）有限公司 | Concentration degree analysis method and device, electronic equipment and computer storage medium |
Non-Patent Citations (3)
| Title |
|---|
| 邓宇辉: "样本区间的概率分布", 《数学杂志》 * |
| 郭克友: "一种计算驾驶人面部朝向角度算法", 《计算机工程与应用》 * |
| 钟佑明 等: "区间概率随机变量及其数字特征", 《重庆大学学报（自然科学版）》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117197853A (en) * | 2022-05-31 | 2023-12-08 | 青岛云天励飞科技有限公司 | Face angle prediction method, device, equipment and readable storage medium |
| CN117197853B (en) * | 2022-05-31 | 2025-10-10 | 青岛云天励飞科技有限公司 | Face angle prediction method, device, equipment and readable storage medium |
| CN116844210A (en) * | 2023-07-04 | 2023-10-03 | 宁夏大学 | Head posture estimation method and system based on structure isodegeneration self-supervision learning |
| WO2025180426A1 (en) * | 2024-02-26 | 2025-09-04 | 北京罗克维尔斯科技有限公司 | Face orientation detection method and apparatus, and device, vehicle, medium and program product |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113034602B (en) | 2023-04-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US7362885B2 (en) | Object tracking and eye state identification method | |
| US9501720B2 (en) | Object detection apparatus | |
| CN101344922B (en) | Human face detection method and device | |
| US7925093B2 (en) | Image recognition apparatus | |
| US9646215B2 (en) | Eye part detection apparatus | |
| US20130070105A1 (en) | Tracking device, tracking method, and computer program product | |
| US20200250803A1 (en) | Method for detecting and tracking target object, target object tracking apparatus, and computer-program product | |
| CN113034602B (en) | Orientation angle analysis method and device, electronic equipment and storage medium | |
| CN111598117B (en) | Image recognition method and device | |
| CN113255557B (en) | Deep learning-based video crowd emotion analysis method and system | |
| CN110969642B (en) | Video filtering method and device, electronic equipment and storage medium | |
| US9489737B2 (en) | Object detection apparatus and method | |
| Ali et al. | Multiple object tracking with partial occlusion handling using salient feature points | |
| US11080562B1 (en) | Key point recognition with uncertainty measurement | |
| CN111860196A (en) | Hand operation action scoring device and method and computer readable storage medium | |
| CN113807119B (en) | Personnel gazing position detection method and device | |
| KR20190050551A (en) | Apparatus and method for recognizing body motion based on depth map information | |
| KR20190009006A (en) | Real time multi-object tracking device and method by using global motion | |
| CN116740792A (en) | Face recognition method and system for sightseeing vehicle operators | |
| CN116189157A (en) | Driver attention heat map generation algorithm and safety prompt system | |
| CN117475358B (en) | Collision prediction method and device based on unmanned aerial vehicle vision | |
| CN108985216B (en) | Pedestrian head detection method based on multivariate logistic regression feature fusion | |
| CN113989914B (en) | Security monitoring method and system based on face recognition | |
| Kim et al. | Driving environment assessment using fusion of in-and out-of-vehicle vision systems | |
| Choi et al. | In and out vision-based driver-interactive assistance system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |